NVlabs/SegFormer

single-gpu training is ok, but Multi-gpu training has some wrong result

Opened this issue · 0 comments

hi, everyone. my single-gpu training is ok, but Multi-gpu training is not successful in evaluaton period. its wrong info:
IndexError: boolean index did not match indexed array along dimension 0; dimension is XXX but corresponding boolean dimension is YYY
XXX and YYY is a test image size
I check my dataset, it is ok, and can be used in other training model, such as deeplab.
so anybody can help me? thanks a lot!